[EM] Withdrawals

Juho Laatu juho.laatu at gmail.com
Thu Nov 17 14:44:45 PST 2016


> On 17 Nov 2016, at 16:20, Jameson Quinn <jameson.quinn at gmail.com> wrote:
> 
> Here again is the scenario I gave:
> 
> 30: AX>B 
> 5: AX>C (accepting C may be purely strategic, or honest; we don't know)
> 15: B>A
> 10: B>AC
> 29: C>B
> 11: CB
> 
> In PAR, C is initially the leader but then loses to A. However, if C withdraws, B becomes the leader and wins.
> 
> In this election, C is actually a spoiler in 2 ways, and close to being a spoiler in 3. 
> Presumably, without C in the race, the AX>B and perhaps also the B>A voters would use ">>"; that is, they would reject instead of accepting. (In fact, I included candidate X in this scenario just so that the B>AC candidates would have someone to reject, so it seems unlikely that the B>A voters would really refrain from rejecting A just so they could make a distinction between A and X.)
That sounds complicated from the voter point of view. Normal voters are not that educated. They would probably follow the strategic advice of the parties / candidates if they want to use such strategies. Not good.
> Also presumably, without C in the race, the C>B voters would top-rate B, so that B would be the clear winner here.
> If the CB voters had taken C more seriously, they might have voted C>B, so that B, despite being the CW, could not be frontrunner even if C withdrew.
> It would be possible to rewrite the PAR rules so that each candidate was tested as a possible frontrunner, and if any candidate could hold that position, the winner would be the one of those with the most "prefer" votes. That would, as Juho suggests, make withdrawal irrelevant/unnecessary. (Though just like withdrawal itself, and unlike FBPPAR, it wouldn't solve the FBC issue completely; changing the CB votes above to C>B, it would still elect A, even though B is the CW, and thus it would motivate favorite betrayal by a small group of C voters.) 
> 
> However, I think that "PAR with automatic withdrawal" would be too complex a system. I don't think that the advantages justify that extra complexity.

Yes. Probably you can achieve what you want also with a simpler system. Maybe one could design a system that is maximally resistant against favourite betrayal by letting the system automatically betray the favourite whenever that is beneficial to the voter. I'm not going to present such a system right now, and I'm sorry I haven't read all the recent mails, so I don't know if someone already did something in that direction :-) . Or maybe one of the well known methods can be justified following this kind of (defensive favourite betrayal) argumentation.

> 
> "PAR with optional withdrawal" is easier to explain, and has most of the advantages of the automatic version. It does not create a scenario where it depends who withdraws in which order. And as for Juho's worry that it would allow C to "sell" the election to B: I think that laws could prevent purely self-interested "sales" like that, and if the "sale" is in return for some ideological tradeoff ("appoint somebody from my party to cabinet post X"), I think that's actually a good thing; if C the level of support it would take to create a scenario like the above, they probably deserve some degree of ideological concessions.

C could sell the election to either A or B. I don't like the idea, and those C supporters that preferred B to A would not be very happy if C would decide that A should win. People might feel that their vote is not worth much, if the end result of the election will be negotiated by the candidates, and not voted by the voters. I'd thus try to avoid giving such strategic power to one person, if possible.

I'm however quite used to that kind of compromise negotiations since in a multiparty country negotiations on the government coalition often have such characteristics. That negotiation takes however place only after the election. I.e. it does not change the voted result of the (parliamentary) election, but only determines the government coalition.

BR, Juho


> 
> 2016-11-17 6:40 GMT-05:00 Juho Laatu <juho.laatu at gmail.com <mailto:juho.laatu at gmail.com>>:
> In the presented example C can say "I will withdraw", which means the same as "never consider me as leader". C can change the winner from A to B.
> 
> The opinion space seems to be such that A and B are quite close to each others, and C is closer to B than to A. In this set-up it may make sense to allow B to win. But some questions remain. Why didn't the method elect B in the first place? Are withdrawals an integral part of the method in the sense that the method does not give ideal results without withdrawals, but only after ideal withdrawals? Should the candidates base their withdrawal decision on their own preferences ("do I prefer B to A") or on the preferences in those voters/ballots that gave them support? Is it a problem that C can sell victory to A, and not withdraw from the race?
> 
> If withdrawals should follow some known rules (e.g. to respect the interest of one's supporters), such rules could be implemented already in the actual method. I mean rules like "if majority of first preference supporters of candidate x would benefit of the withdrawal, then candidate x shall be automatically considered withdrawn". There may be some problems with simultaneous withdrawals, or the order of them, in both automatic and candidate decision based withdrawal scenarios.
> 
> BR, Juho
> 
> 
> 
> PAR description: 
> Voters Prefer, Accept, or Reject each candidate. On ballots which don't explicitly use "Reject", or for candidates with less than 25% "Prefer", blanks count as "Reject"; otherwise, blanks count as "Accept".
> Tally 1 point for each "Prefer" for each candidate.
> Out of the candidates (if any) with no more than 50% "Reject", find the one with the most points. For every ballot which doesn't "Prefer" this frontrunner, add 1 point for each "Accept".
> If the frontrunner still has the most points, they win. Otherwise, the winner is the candidate with fewest "Reject" ratings.
> 
> 
> 
>> On 16 Nov 2016, at 16:21, Jameson Quinn <jameson.quinn at gmail.com <mailto:jameson.quinn at gmail.com>> wrote:
>> 
>> In PAR, withdrawing makes sense in the many of the same cases where favorite betrayal does. For instance:
>> 
>> 30: AX>B 
>> 5: AX>C (accepting C may be purely strategic, or honest; we don't know)
>> 15: B>A
>> 10: B>AC
>> 29: C>B
>> 11: CB
>> 
>> C is initially considered the "frontrunner", but cannot hold that lead against A; then A wins as the least-rejected. But if C were rejected, B would the initial frontrunner, then that lead would hold and B would win; since C's top-voters universally prefer that result over A winning, presumably C would want to withdraw.
>> 
>> This is the same situation where, in FBPPAR, C's top-voters would be strategically advised to mark C as "stand aside".
>> 
>> Note that, unlike the cases Juho pointed out, this does not break down in the case of a Condorcet cycle. "Withdraw" would not mean "don't let me win", it would just mean "never consider me as leader"; so in a CC, all candidates would rationally withdraw, and PAL would simply devolve to approval using all above-bottom votes. (Note that this is the rational equilibrium vote in approval for all the pro-cyclical voters, by definition a majority, and the locally-optimal-though-not-equilibrium vote for a substantial fraction of the anti-cyclical voters.)
>> 
>> I think that withdrawal is easier to explain than FBPPAR, so I think PAR-with-withdrawal is a good option if you want a system that's even closer to FBC than PAR. 
>> 
>> 2016-11-16 5:11 GMT-05:00 Juho Laatu <juho.laatu at gmail.com <mailto:juho.laatu at gmail.com>>:
>> Hi guys. I have not read the whole stream but I throw in some words on withdrawals anyway.
>> 
>> One must accept that the winner can withdraw since it is always possible that the winner becomes sick and we can do nothing about that. It is impossible to tell if he is withdrawing for this kind of reasons or for some other less acceptable reasons (e.g. if some of the competitors pays him money to withdraw, or if he took part in the election just for fun to show that he can beat the others).
>> 
>> Allowing others to withdraw is more problematic. Let's say there's a Condorcet cycle A>B>C>A (in some Condorcet method), and one of the candidates (B) can change the winner (A) by withdrawing from the race. This would make it possible for him to sell the victory to one of the competitors (C in this example) of the (original, quite likely sincere) winner. That is probably not what we want.
>> 
>> In Condorcet methods it is possible that many of the B voters would like the idea of changing the winner from A to C (they may have voted B>C>A). But probably not all. If we allow withdrawals, and they are commonly used, maybe (in a three candidate cycle A>B>C>A) the method should declare B the original winner instead of A, since that would make it possible for C to withdraw, and A would win (the candidate that should have won in our opinion in the first place). That doesn't sound good. Maybe it is better to use the election method to pick the best winner, and not allow the candidates to agree after the election who the winner is.
>> 
>> I note that we could allow any of the candidates to "withdraw conditionally if they win". In that case the withdrawal of a non-winning candidate would never change the winner. The votes would stay the same even after this kind of withdrawals. This would however allow the U.S. presidential election style withdrawals where the other candidate (or all the other candidates?) can withdraw when they are sure that they have no chances to win. This kind of withdrawal could change the winner only in the case that a recount would make the withdrawing candidate the winner. And in this case we would back in the scenario where the winner is in any case allowed to withdraw.
>> 
>> Some methods might benefit of the possibility of allowing a candidate to withdraw in order to allow a clone candidate to win. If the method has this kind of serious problems with clone nomination, it might be better to use some candidate declared strategy instead of allowing individual candidates to change the result by allowing them to strategically withdraw. A simple declared strategy (in some methods) could be to allow those candidates to declare themselves as clones _before_ the election. In the cyclic ranked example (A>B>C>A), B and C could be candidates of the same party, and it would make sense to allow one of them to win, if party "BC" has majority. Even if A would be the best winner with sincere votes, we may agree that B (or maybe C) is the best winner (with about 66% majority). We probably had quite a few C>A>B voters that would want A to win instead of B. But they knew already when they voted for C that B would inherit those votes. Fair deal, no betrayal, nor strategic plotting after the election.
>> 
>> (An extreme method with this kind of declared candidate strategy would be one where the candidates are organized as a complete binary tree. It would not make sense to rank the candidates. Bullet voting would be sufficient since the inheritance route of all the votes would be fixed anyway.)
>> 
>> BR, Juho
>> 
>> 
>> 
>>> On 16 Nov 2016, at 03:21, Michael Ossipoff <email9648742 at gmail.com <mailto:email9648742 at gmail.com>> wrote:
>>> 
>>> Yes, if it's freely negotiated by the candidates, then there wouldn't be that vulnerability to strategic voters.
>>> 
>>> Michael Ossipoff
>>> 
>>> On Tue, Nov 15, 2016 at 7:39 PM, Forest Simmons <fsimmons at pcc.edu <mailto:fsimmons at pcc.edu>> wrote:
>>> It's not really clear to me; it seems like it depends on the dynamics of the candidate withdrawals when no candidate is forced to withdraw until they think it is safe to do so.  There might be some bluffing and standoffs, so it would depend on the negotiation skills of the candidates.  Especially if they are allowed to decide on the fly whom to transfer the unspecified choices to.
>>> 
>>> On Tue, Nov 15, 2016 at 2:58 PM, Michael Ossipoff <email9648742 at gmail.com <mailto:email9648742 at gmail.com>> wrote:
>>> But, if that will elect the CWs, with sincere voting, then isn't there still an incentive and vulnerability for some faction to falsify preferences to make a strategic cycle that they might win, if they don't like the likely CWs?
>>> 
>>> Michael Ossipoff
>>> 
>>> On Tue, Nov 15, 2016 at 3:10 PM, Forest Simmons <fsimmons at pcc.edu <mailto:fsimmons at pcc.edu>> wrote:
>>> Combining Candidate Withdrawal with Candidate Transfer:
>>> 
>>> When a candidate withdraws, hir votes are distributed in accordance with the additional preferences of hir supporters.  When no additional preferences are specified, it is assumed that the voter supports the preference order of the withdrawing candidate.
>>> 
>>> Voters may place NOTB (none of the below) above their truncated candidates.  If at any stage the NOTB candidate accumulates more than two-thirds (say) of the votes, the election is scuttled and new candidates must be nominated.
>>> 
>>> On Mon, Nov 14, 2016 at 9:55 PM, Michael Ossipoff <email9648742 at gmail.com <mailto:email9648742 at gmail.com>> wrote:
>>> But, even if that rule were extended to rescue a CWv that's farther down in the rankings  I guess that would just be like having a rule to elect the CWv if there is one, and otherwise to do IRV, which would be a less-good relative of Benham & Woodall.
>>> 
>>> And of course a rule to elect  the CWv is always vulnerable to strategic falsification of preferences to make a strategic cycle, with the problems that go with that.
>>> 
>>> In fact, I guess that the candidate-withdrawal option, Steve Eppley's original proposal, would, as well, be vulnerable to a faction making a strategic cycle, to avoid a result in which the election of the CWv is inevitable due to the candidate-withdrawal that will happen if ordinary IRV doesn't elect the CWv..
>>> 
>>> Oh well, so much for that possibility :^)
>>> 
>>> Michael Ossipoff
>>> 
>>> On Mon, Nov 14, 2016 at 10:47 PM, Michael Ossipoff <email9648742 at gmail.com <mailto:email9648742 at gmail.com>> wrote:
>>> Speaking of candidate-participation methods, Steve Eppley proposed the simplest one: the Candidate Withdrawal option.
>>> 
>>> When the count result is announced, any candidate has 24 hours in which to announce hir withdrawal from the election, and call for a re-count with hir deleted from the ballots.
>>> 
>>> An IRVist objected to that, because withdrawal betrays the candidate's voters.  Ok, then add a rule that a winning candidate can't withdraw :^)
>>> 
>>> The candidate withdrawal option would completely save IRV from its problem. Candidate-withdrawal should be offered as an amendment to IRV proposals, and also to already-enacted IRV.
>>> 
>>> But, just looking at IRV, the circumstances for wanting to withdraw (Your voters let the CWv get eliminated, by ranking hir 2nd instead of 1st, and hir votes transferred the other way) seem so well-defined and specific as to suggested an automated candidate-withdrawal, built into the IRV count-rule.
>>> 
>>> How about saying that, when that happens, the count rule will delete the 1st choice that is in the way of the eliminated CWv being at the top of the rankings that didn't help CWv because they ranked hir 2nd, and their vote never got to hir?   ...followed by a 2nd count without the deleted 1st choice?
>>> 
>>> ...or it could be an option for the voter to choose if s/he wants that deletion and re-count to occur if hir 1st choice eliminates a CWv, resulting the election of someone over whom you ranked the CWv.   ...where, without that 1st choice, the CWv would have won.
>>> 
>>> That option would only apply to the first election result, and not to subsequent ones.
>>> 
>>> The method could be called "Automatic Strategic Dropping"
>>> 
>>> Or, IRV with that option could be called "Auto-Revotable IRV".
>>> 
>>> I don't like the name IRV. Its meaning isn't a good description. RCV is even worse, because _every_ rank-balloting method is RCV.
>>> 
>>> I'd like to rename IRV as "Delete Least Favorite", "Delete Least Top", "Eliminate Least Favorite" or "Eliminate Least Top".
>>> 
>>> That would be DLF, DLT, ELF, or ELT.  How about ELF?
>>> 
>>> That name goes with my brief definition of the IRV count:
>>> 
>>> Repeatedly, delete from the rankings the candidate currently topping fewest rankings.
>>> 
>>> [end of definition of ELF definition]
>>> 
>>> If that action is repeated, it will eventually leave only one un-deleted candidate.
>>> 
>>> So "Auto-Revotable Eliminate Least Favorite"?
>>> 
>>> Michael Ossipoff
>>> 
>>> 
>>> 
>>> 
>>> 
>>> On Mon, Nov 14, 2016 at 6:57 PM, Forest Simmons <fsimmons at pcc.edu <mailto:fsimmons at pcc.edu>> wrote:
>>> 
>>> 
>>> On Sat, Nov 12, 2016 at 9:29 PM, Michael Ossipoff <email9648742 at gmail.com <mailto:email9648742 at gmail.com>> wrote:
>>> Yes, it would make sense, and be interesting for a voting-system reform proposal to include CVT in its list of offered methods. Offer a complete variety, and let the initiative committee &/or the public choose. The addition of CVT would greatly broaden the variety. So the list could be:
>>> 
>>> Approval
>>> Score
>>> Bucklin
>>> MDDTR
>>> CVT
>>> 
>>> Actually, it might be best to include IRV, with a fair assessment of its advantages & disadvantages, so that people won't think that IRV is being prejudcially ignored, and so as to answer, in advance, any arguments from FairVote.
>>> 
>>> Michael Ossipoff
>>> 
>>> I agree, IRV should be included for the reasons you suggest.
>>> 
>>> Also, I think that the inclusion of a NOTA (none of the above) option on the ballot should be a part of the CTV suggestion:  If NOTA gets the highest total it means that too few of the voters trust the competence and integrity of their favorites to act as proxies for them.
>>> 
>>> 
>>> 
>>> 
>>> 
>>> 
>>> ----
>>> Election-Methods mailing list - see http://electorama.com/em <http://electorama.com/em> for list info
>> 
>> 
>> ----
>> Election-Methods mailing list - see http://electorama.com/em <http://electorama.com/em> for list info
>> 
>> 
> 
> 
> ----
> Election-Methods mailing list - see http://electorama.com/em <http://electorama.com/em> for list info
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20161118/a12dae9a/attachment-0001.htm>


More information about the Election-Methods mailing list